In order to maximise profits, avoid stockouts and excess inventory, and keep demand and supply in equilibrium, demand forecasting is an essential component of supply chain management for manufacturers, wholesalers, and trading companies. Underestimating demand causes supply chain inefficiencies like missed sales opportunities, poor service levels, and unfulfilled orders; overestimating demand causes storage costs to rise and surplus inventory to accumulate. This research suggests a three-stage forecasting approach that includes preprocessing data, extracting features, and training the model to overcome these obstacles. Preprocessing is a smoothing technique to improve data quality, and feature extraction uses a mix of a priori and probabilistic methods to find meaningful patterns. Fundamentally new is a recommendation model that combines the CatBoost algorithm with a DCN to enhance prediction accuracy and cater to supply chain users' individual requirements. With an accuracy of up to 96.61%, the suggested DCN-CatBoost hybrid model far surpasses current techniques, according to comparative evaluations. These outcomes demonstrate how well the model works to improve the accuracy of forecasts, which in turn helps supply chain operations be more responsive and efficient. In order to improve SCM Demand Forecasting, this study highlights the use of sophisticated analytics. © 2025 IEEE.